Enhancing Machine Learning and Data Visualization Pipelines with Isomorphisms
summary
Claims that realizing isomorphisms inside pipelines reduces unnecessary transformations and raises explainability while not magically fixing algorithmic cost. Implements a relational‑algebra k‑nearest‑neighbors on Fisher's Iris (Euclidean distance, 4 dimensions, 150 rows → ~90,000 pairwise ops), stores data in sixth normal form, prototypes in a Julia‑based logic DB, compares notes with mlpack work on billion‑row sets, and emits results to Vega‑Lite for labeled visualization.
AI-written summary · about
chapters (11)
- 0:00Start
- 3:44Graph Isomorphism
- 4:54Bijection
- 10:57Fischer's Iris Data Set
- 13:35Relational View (6NF)
- 35:38End-to-End Operation
- 46:32data out → concat
- 54:18Finished Operation
- 57:37Enhancements
- 58:23References
- 58:39Appendix Euclid distance for tuples of airity n
related talks
27mScalable Data Pipelines - an ML Approach
34mInterpreting Machine Learning Models: Why and How!
24mHybrid ML Methods for the Interpretation and Integration of Heterogeneous Multimodal Data
35mWriting Machine Learning Pipelines Against Object Storage
33mBuilding Resilient (ML) Pipelines for MLOps
01h 40mBuilding machine learning pipelines that scale
40mTools and Tricks from a Pragmatic Data Scientist
01h 21mBuilding a ML Model Observability pipeline